Image compression method and apparatus which satisfies a predefined bit budget

1. A method of compressing a digitised image composed of a matrix of image samples to provide a compressed image which satisfies a predefined bit budget, the method comprising the steps of: 1) dividing the digitised image into blocks and deriving for each block an energy packing transform comprising a set of transform coefficients; 2) selecting a quantisation table from a set of quantisation tables and using the selected table to quantise the coefficients of each transform; 3) deriving a zero-value index indicative of the number of zero-value quantised transform coefficients; 4) determining a predicted zero-value index using said predefined bit budget; 5) selecting a quantisation table from said set of tables using the derived index and said predicted index and using that selected table to quantise the coefficients of each transform; and 6) compressing the coefficients quantised in step 5) using run-length encoding.

2. A method according to claim 1, wherein said zero-value index is the average number of zero-value quantised transform coefficients per transform.

3. A method according to claim 1, wherein step 3) comprises deriving zero-value indices for each of a plurality of different quantisation tables to provide a zero-value index vs quantisation table relationship.

4. A method according to claim 1 and comprising obtaining a reference zero-value index vs bit budget relationship by: 7) dividing a digitised test image into blocks and deriving for each block an energy packing transform comprising a set of transform coefficients; 8) selecting a quantisation table from a set of quantisation tables and using the selected table to quantise the coefficients of each transform; 9) deriving a zero-value index indicative of the number of zero-value quantised transform coefficients; 10) compressing the coefficients quantised in step 5) using run-length encoding; 11) determining the bit size of the compressed image; 12) repeating steps 7) to 11) for a plurality of different quantisation tables to obtain a zero-value vs bit size relationship for the test image; and 13) repeating steps 7) to 12) for a plurality of different test images and combining the resulting relationships to obtain the reference zero-value index vs bit budget relationship, wherein this relationship is used in step 4) to determine the predicted zero-value index using the predefined bit budget.

5. A method according to claim 4, wherein step 5) comprises using the predicted zero-value index and the zero-value index vs quantisation table relationship for the image to be compressed, to select the quantisation table.

6. A method according to claim 1, wherein said energy packing transform is a discrete cosine transform (DCT).

7. A method according to claim 1, wherein step 6) comprises entropy encoding the data after run-length encoding.

8. Apparatus for compressing a digitised image composed of a matrix of image samples to provide a compressed image which satisfies a predefined bit budget, the apparatus comprising: first signal processing means for dividing the digitised image into blocks and for deriving for each block an energy packing transform comprising a set of transform coefficient; a quantisation table specification memory storing a set of quantisation tables; quantisation means for quantising the coefficients of each transform using a first quantisation table selected from said set of quantisation tables; second signal processing means for deriving an index representative of the number of zero-value quantised transform coefficients, for determining a predicted zero-value index using said predefined bit budget, for selecting a quantisation table from said set of tables using the derived index and said predicted index and using that selected table to quantise the coefficients of each transform; and encoding means for compressing the coefficients quantised by the second signal processing means using run-length encoding.